Google DeepMind open-sources Science Skills for AI research workflows
DeepMind released Science Skills, a Python library of modular functions that let language models call scientific tools, databases, and simulations as part of reproducible research workflows.
Google DeepMind released Science Skills this week, a framework for building reproducible AI research workflows. The project, now live on GitHub, provides a collection of Python-based "skills" — modular functions that language models can call to interact with scientific tools, datasets, and computational environments. The goal is to move beyond one-off research scripts toward a shared library of validated, reusable components.
Science Skills targets researchers building agents that need to run simulations, query databases, parse literature, or orchestrate multi-step experiments. Each skill is a self-contained Python function with explicit inputs, outputs, and error handling. The framework includes example skills for common tasks: fetching PubMed abstracts, running molecular dynamics simulations, querying protein structure databases, and formatting results for downstream analysis. DeepMind positions the project as infrastructure for the next generation of AI-assisted science, where models don't just generate text but actively participate in the research loop.
What stands out
- 01Standardized interfaces. Every skill follows a consistent signature: typed inputs, structured outputs, and machine-readable error codes. This makes it easier to chain skills together or swap implementations without rewriting calling code.
- 02Validation and testing. The repo includes unit tests and integration tests for each skill. DeepMind's internal teams have been using these skills in production research projects, so the library reflects real-world debugging and edge-case handling.
- 03Domain coverage. Initial skills span biology (UniProt queries, PDB lookups), chemistry (RDKit wrappers, reaction enumeration), and physics (simulation launchers for GROMACS and LAMMPS). The framework is designed to be extensible — researchers can contribute new skills via pull request.
- 04




